SLI in reverse
- Thread starter Frank
- Start date
I think CPUs are still the primary bound for quite a few recent games. HL2 in particular seems very CPU-sensitive, looking at Rollo's 6800 SLI vs. 6800GT SLI benches at AT.
SLI offers noticable boosts at 16x12 4x16 in current games, but I think anything beyond that (ignoring other constraints, like inter-board communication) is probably CPU-limited for current games.
Well, Riddick's PS2++ and Far Cry's HDR modes may appreciate the extra pipes, but the current SLI implementation apparently won't help that.
SLI offers noticable boosts at 16x12 4x16 in current games, but I think anything beyond that (ignoring other constraints, like inter-board communication) is probably CPU-limited for current games.
Well, Riddick's PS2++ and Far Cry's HDR modes may appreciate the extra pipes, but the current SLI implementation apparently won't help that.
This is going to come off as possibly a bit of a stupid question, but I have to ask because, well I have a project. Also sorry in advance, I'm not certain if this would be a pertinant question for this thread.
1.) Has anybody any experience with massively distributed realtime rendering? I.E. has any body any experience with grouping more than two or four modern GPUs (via separate systems... accounting for the knowledge that bandwidth is a severe limitation) for the purpose of rendering large detail scenes, of a few million polygons for art assetts in realtime?
2.) If not, does anybody know of any project with any degree of similarity to this (excluding SGI Prism), or is it without precedent?
3.) If so, are there any sources that can be researched upon?
The project involves using off the shelf hardware for the purpose of advanced visualisation. Think something similar to an Nvidia/ATI tech demo, only more complex with actual environmental interraction and simulation. I'm near certain that due to bandwidth limitations the entire project will have to be scrapped, and reworked.
Any help would be greatly appreciated.
Thanks
1.) Has anybody any experience with massively distributed realtime rendering? I.E. has any body any experience with grouping more than two or four modern GPUs (via separate systems... accounting for the knowledge that bandwidth is a severe limitation) for the purpose of rendering large detail scenes, of a few million polygons for art assetts in realtime?
2.) If not, does anybody know of any project with any degree of similarity to this (excluding SGI Prism), or is it without precedent?
3.) If so, are there any sources that can be researched upon?
The project involves using off the shelf hardware for the purpose of advanced visualisation. Think something similar to an Nvidia/ATI tech demo, only more complex with actual environmental interraction and simulation. I'm near certain that due to bandwidth limitations the entire project will have to be scrapped, and reworked.
Any help would be greatly appreciated.
Thanks
Similar threads
